The More Alerts, the Fewer Get Read
There is a bot that watches the stock sheet of a small trading company. Its first rule was simple: 「alert when remaining quantity is 10 or fewer」. The result was 14 alerts a day, and by the third day I had stopped reading the pile that built up in Slack each morning.
One Unit Changed
I replaced quantity with 「estimated days left」. The bot divides the remaining stock by the average daily shipping speed of the last two weeks, and only alerts when the result is 7 days or less.
Before: quantity <= 10 → alert
After : quantity ÷ daily shipments <= 7 days → alert
Before / After
• **Stock alerts**: 14 a day → 3
• **Buyer lead collection**: the same names showed up repeatedly and I cross-checked by hand → duplicates are removed by domain first, and I only confirm
• **Email drafts**: restock notice drafts are written only for items that triggered an alert, and nothing is sent until I read and approve it
An Unexpected Finding
With fewer alerts, response time actually improved. Three alerts get read to the end, and what gets read gets acted on. Ten units of a slow item was not urgent, while fifteen units of a fast item was a real warning. The same number means nothing without speed, and I realized that later than the bot did.
Today's Notes
• Base alerts on speed, not just quantity
• Shrink the volume to what a person can finish reading
• Keep the send button in human hands until the very end